2 * net/sched/act_api.c Packet action API.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Author: Jamal Hadi Salim
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/string.h>
17 #include <linux/errno.h>
18 #include <linux/slab.h>
19 #include <linux/skbuff.h>
20 #include <linux/init.h>
21 #include <linux/kmod.h>
22 #include <linux/err.h>
23 #include <linux/module.h>
24 #include <linux/rhashtable.h>
25 #include <linux/list.h>
26 #include <net/net_namespace.h>
28 #include <net/sch_generic.h>
29 #include <net/pkt_cls.h>
30 #include <net/act_api.h>
31 #include <net/netlink.h>
33 static int tcf_action_goto_chain_init(struct tc_action *a, struct tcf_proto *tp)
35 u32 chain_index = a->tcfa_action & TC_ACT_EXT_VAL_MASK;
39 a->goto_chain = tcf_chain_get_by_act(tp->chain->block, chain_index);
45 static void tcf_action_goto_chain_fini(struct tc_action *a)
47 tcf_chain_put_by_act(a->goto_chain);
50 static void tcf_action_goto_chain_exec(const struct tc_action *a,
51 struct tcf_result *res)
53 const struct tcf_chain *chain = a->goto_chain;
55 res->goto_tp = rcu_dereference_bh(chain->filter_chain);
58 static void tcf_free_cookie_rcu(struct rcu_head *p)
60 struct tc_cookie *cookie = container_of(p, struct tc_cookie, rcu);
66 static void tcf_set_action_cookie(struct tc_cookie __rcu **old_cookie,
67 struct tc_cookie *new_cookie)
69 struct tc_cookie *old;
71 old = xchg((__force struct tc_cookie **)old_cookie, new_cookie);
73 call_rcu(&old->rcu, tcf_free_cookie_rcu);
76 /* XXX: For standalone actions, we don't need a RCU grace period either, because
77 * actions are always connected to filters and filters are already destroyed in
78 * RCU callbacks, so after a RCU grace period actions are already disconnected
79 * from filters. Readers later can not find us.
81 static void free_tcf(struct tc_action *p)
83 free_percpu(p->cpu_bstats);
84 free_percpu(p->cpu_qstats);
86 tcf_set_action_cookie(&p->act_cookie, NULL);
88 tcf_action_goto_chain_fini(p);
93 static void tcf_action_cleanup(struct tc_action *p)
98 gen_kill_estimator(&p->tcfa_rate_est);
102 static int __tcf_action_put(struct tc_action *p, bool bind)
104 struct tcf_idrinfo *idrinfo = p->idrinfo;
106 if (refcount_dec_and_lock(&p->tcfa_refcnt, &idrinfo->lock)) {
108 atomic_dec(&p->tcfa_bindcnt);
109 idr_remove(&idrinfo->action_idr, p->tcfa_index);
110 spin_unlock(&idrinfo->lock);
112 tcf_action_cleanup(p);
117 atomic_dec(&p->tcfa_bindcnt);
122 int __tcf_idr_release(struct tc_action *p, bool bind, bool strict)
126 /* Release with strict==1 and bind==0 is only called through act API
127 * interface (classifiers always bind). Only case when action with
128 * positive reference count and zero bind count can exist is when it was
129 * also created with act API (unbinding last classifier will destroy the
130 * action if it was created by classifier). So only case when bind count
131 * can be changed after initial check is when unbound action is
132 * destroyed by act API while classifier binds to action with same id
133 * concurrently. This result either creation of new action(same behavior
134 * as before), or reusing existing action if concurrent process
135 * increments reference count before action is deleted. Both scenarios
139 if (!bind && strict && atomic_read(&p->tcfa_bindcnt) > 0)
142 if (__tcf_action_put(p, bind))
148 EXPORT_SYMBOL(__tcf_idr_release);
150 static size_t tcf_action_shared_attrs_size(const struct tc_action *act)
152 struct tc_cookie *act_cookie;
156 act_cookie = rcu_dereference(act->act_cookie);
159 cookie_len = nla_total_size(act_cookie->len);
162 return nla_total_size(0) /* action number nested */
163 + nla_total_size(IFNAMSIZ) /* TCA_ACT_KIND */
164 + cookie_len /* TCA_ACT_COOKIE */
165 + nla_total_size(0) /* TCA_ACT_STATS nested */
166 /* TCA_STATS_BASIC */
167 + nla_total_size_64bit(sizeof(struct gnet_stats_basic))
168 /* TCA_STATS_QUEUE */
169 + nla_total_size_64bit(sizeof(struct gnet_stats_queue))
170 + nla_total_size(0) /* TCA_OPTIONS nested */
171 + nla_total_size(sizeof(struct tcf_t)); /* TCA_GACT_TM */
174 static size_t tcf_action_full_attrs_size(size_t sz)
176 return NLMSG_HDRLEN /* struct nlmsghdr */
177 + sizeof(struct tcamsg)
178 + nla_total_size(0) /* TCA_ACT_TAB nested */
182 static size_t tcf_action_fill_size(const struct tc_action *act)
184 size_t sz = tcf_action_shared_attrs_size(act);
186 if (act->ops->get_fill_size)
187 return act->ops->get_fill_size(act) + sz;
191 static int tcf_dump_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
192 struct netlink_callback *cb)
194 int err = 0, index = -1, s_i = 0, n_i = 0;
195 u32 act_flags = cb->args[2];
196 unsigned long jiffy_since = cb->args[3];
198 struct idr *idr = &idrinfo->action_idr;
200 unsigned long id = 1;
202 spin_lock(&idrinfo->lock);
206 idr_for_each_entry_ul(idr, p, id) {
212 time_after(jiffy_since,
213 (unsigned long)p->tcfa_tm.lastuse))
216 nest = nla_nest_start(skb, n_i);
219 goto nla_put_failure;
221 err = tcf_action_dump_1(skb, p, 0, 0);
224 nlmsg_trim(skb, nest);
227 nla_nest_end(skb, nest);
229 if (!(act_flags & TCA_FLAG_LARGE_DUMP_ON) &&
230 n_i >= TCA_ACT_MAX_PRIO)
235 cb->args[0] = index + 1;
237 spin_unlock(&idrinfo->lock);
239 if (act_flags & TCA_FLAG_LARGE_DUMP_ON)
245 nla_nest_cancel(skb, nest);
249 static int tcf_del_walker(struct tcf_idrinfo *idrinfo, struct sk_buff *skb,
250 const struct tc_action_ops *ops)
255 struct idr *idr = &idrinfo->action_idr;
257 unsigned long id = 1;
259 nest = nla_nest_start(skb, 0);
261 goto nla_put_failure;
262 if (nla_put_string(skb, TCA_KIND, ops->kind))
263 goto nla_put_failure;
265 idr_for_each_entry_ul(idr, p, id) {
266 ret = __tcf_idr_release(p, false, true);
267 if (ret == ACT_P_DELETED) {
268 module_put(ops->owner);
270 } else if (ret < 0) {
271 goto nla_put_failure;
274 if (nla_put_u32(skb, TCA_FCNT, n_i))
275 goto nla_put_failure;
276 nla_nest_end(skb, nest);
280 nla_nest_cancel(skb, nest);
284 int tcf_generic_walker(struct tc_action_net *tn, struct sk_buff *skb,
285 struct netlink_callback *cb, int type,
286 const struct tc_action_ops *ops,
287 struct netlink_ext_ack *extack)
289 struct tcf_idrinfo *idrinfo = tn->idrinfo;
291 if (type == RTM_DELACTION) {
292 return tcf_del_walker(idrinfo, skb, ops);
293 } else if (type == RTM_GETACTION) {
294 return tcf_dump_walker(idrinfo, skb, cb);
296 WARN(1, "tcf_generic_walker: unknown command %d\n", type);
297 NL_SET_ERR_MSG(extack, "tcf_generic_walker: unknown command");
301 EXPORT_SYMBOL(tcf_generic_walker);
303 static bool __tcf_idr_check(struct tc_action_net *tn, u32 index,
304 struct tc_action **a, int bind)
306 struct tcf_idrinfo *idrinfo = tn->idrinfo;
309 spin_lock(&idrinfo->lock);
310 p = idr_find(&idrinfo->action_idr, index);
314 refcount_inc(&p->tcfa_refcnt);
316 atomic_inc(&p->tcfa_bindcnt);
318 spin_unlock(&idrinfo->lock);
327 int tcf_idr_search(struct tc_action_net *tn, struct tc_action **a, u32 index)
329 return __tcf_idr_check(tn, index, a, 0);
331 EXPORT_SYMBOL(tcf_idr_search);
333 bool tcf_idr_check(struct tc_action_net *tn, u32 index, struct tc_action **a,
336 return __tcf_idr_check(tn, index, a, bind);
338 EXPORT_SYMBOL(tcf_idr_check);
340 int tcf_idr_delete_index(struct tc_action_net *tn, u32 index)
342 struct tcf_idrinfo *idrinfo = tn->idrinfo;
346 spin_lock(&idrinfo->lock);
347 p = idr_find(&idrinfo->action_idr, index);
349 spin_unlock(&idrinfo->lock);
353 if (!atomic_read(&p->tcfa_bindcnt)) {
354 if (refcount_dec_and_test(&p->tcfa_refcnt)) {
355 struct module *owner = p->ops->owner;
357 WARN_ON(p != idr_remove(&idrinfo->action_idr,
359 spin_unlock(&idrinfo->lock);
361 tcf_action_cleanup(p);
370 spin_unlock(&idrinfo->lock);
373 EXPORT_SYMBOL(tcf_idr_delete_index);
375 int tcf_idr_create(struct tc_action_net *tn, u32 index, struct nlattr *est,
376 struct tc_action **a, const struct tc_action_ops *ops,
377 int bind, bool cpustats)
379 struct tc_action *p = kzalloc(ops->size, GFP_KERNEL);
380 struct tcf_idrinfo *idrinfo = tn->idrinfo;
385 refcount_set(&p->tcfa_refcnt, 1);
387 atomic_set(&p->tcfa_bindcnt, 1);
390 p->cpu_bstats = netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
393 p->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
397 spin_lock_init(&p->tcfa_lock);
398 p->tcfa_index = index;
399 p->tcfa_tm.install = jiffies;
400 p->tcfa_tm.lastuse = jiffies;
401 p->tcfa_tm.firstuse = 0;
403 err = gen_new_estimator(&p->tcfa_bstats, p->cpu_bstats,
405 &p->tcfa_lock, NULL, est);
410 p->idrinfo = idrinfo;
412 INIT_LIST_HEAD(&p->list);
416 free_percpu(p->cpu_qstats);
418 free_percpu(p->cpu_bstats);
423 EXPORT_SYMBOL(tcf_idr_create);
425 void tcf_idr_insert(struct tc_action_net *tn, struct tc_action *a)
427 struct tcf_idrinfo *idrinfo = tn->idrinfo;
429 spin_lock(&idrinfo->lock);
430 /* Replace ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
431 WARN_ON(!IS_ERR(idr_replace(&idrinfo->action_idr, a, a->tcfa_index)));
432 spin_unlock(&idrinfo->lock);
434 EXPORT_SYMBOL(tcf_idr_insert);
436 /* Cleanup idr index that was allocated but not initialized. */
438 void tcf_idr_cleanup(struct tc_action_net *tn, u32 index)
440 struct tcf_idrinfo *idrinfo = tn->idrinfo;
442 spin_lock(&idrinfo->lock);
443 /* Remove ERR_PTR(-EBUSY) allocated by tcf_idr_check_alloc */
444 WARN_ON(!IS_ERR(idr_remove(&idrinfo->action_idr, index)));
445 spin_unlock(&idrinfo->lock);
447 EXPORT_SYMBOL(tcf_idr_cleanup);
449 /* Check if action with specified index exists. If actions is found, increments
450 * its reference and bind counters, and return 1. Otherwise insert temporary
451 * error pointer (to prevent concurrent users from inserting actions with same
452 * index) and return 0.
455 int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
456 struct tc_action **a, int bind)
458 struct tcf_idrinfo *idrinfo = tn->idrinfo;
463 spin_lock(&idrinfo->lock);
465 p = idr_find(&idrinfo->action_idr, *index);
467 /* This means that another process allocated
468 * index but did not assign the pointer yet.
470 spin_unlock(&idrinfo->lock);
475 refcount_inc(&p->tcfa_refcnt);
477 atomic_inc(&p->tcfa_bindcnt);
482 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
485 idr_replace(&idrinfo->action_idr,
486 ERR_PTR(-EBUSY), *index);
491 ret = idr_alloc_u32(&idrinfo->action_idr, NULL, index,
492 UINT_MAX, GFP_ATOMIC);
494 idr_replace(&idrinfo->action_idr, ERR_PTR(-EBUSY),
497 spin_unlock(&idrinfo->lock);
500 EXPORT_SYMBOL(tcf_idr_check_alloc);
502 void tcf_idrinfo_destroy(const struct tc_action_ops *ops,
503 struct tcf_idrinfo *idrinfo)
505 struct idr *idr = &idrinfo->action_idr;
508 unsigned long id = 1;
510 idr_for_each_entry_ul(idr, p, id) {
511 ret = __tcf_idr_release(p, false, true);
512 if (ret == ACT_P_DELETED)
513 module_put(ops->owner);
517 idr_destroy(&idrinfo->action_idr);
519 EXPORT_SYMBOL(tcf_idrinfo_destroy);
521 static LIST_HEAD(act_base);
522 static DEFINE_RWLOCK(act_mod_lock);
524 int tcf_register_action(struct tc_action_ops *act,
525 struct pernet_operations *ops)
527 struct tc_action_ops *a;
530 if (!act->act || !act->dump || !act->init || !act->walk || !act->lookup)
533 /* We have to register pernet ops before making the action ops visible,
534 * otherwise tcf_action_init_1() could get a partially initialized
537 ret = register_pernet_subsys(ops);
541 write_lock(&act_mod_lock);
542 list_for_each_entry(a, &act_base, head) {
543 if (act->type == a->type || (strcmp(act->kind, a->kind) == 0)) {
544 write_unlock(&act_mod_lock);
545 unregister_pernet_subsys(ops);
549 list_add_tail(&act->head, &act_base);
550 write_unlock(&act_mod_lock);
554 EXPORT_SYMBOL(tcf_register_action);
556 int tcf_unregister_action(struct tc_action_ops *act,
557 struct pernet_operations *ops)
559 struct tc_action_ops *a;
562 write_lock(&act_mod_lock);
563 list_for_each_entry(a, &act_base, head) {
565 list_del(&act->head);
570 write_unlock(&act_mod_lock);
572 unregister_pernet_subsys(ops);
575 EXPORT_SYMBOL(tcf_unregister_action);
578 static struct tc_action_ops *tc_lookup_action_n(char *kind)
580 struct tc_action_ops *a, *res = NULL;
583 read_lock(&act_mod_lock);
584 list_for_each_entry(a, &act_base, head) {
585 if (strcmp(kind, a->kind) == 0) {
586 if (try_module_get(a->owner))
591 read_unlock(&act_mod_lock);
596 /* lookup by nlattr */
597 static struct tc_action_ops *tc_lookup_action(struct nlattr *kind)
599 struct tc_action_ops *a, *res = NULL;
602 read_lock(&act_mod_lock);
603 list_for_each_entry(a, &act_base, head) {
604 if (nla_strcmp(kind, a->kind) == 0) {
605 if (try_module_get(a->owner))
610 read_unlock(&act_mod_lock);
615 /*TCA_ACT_MAX_PRIO is 32, there count upto 32 */
616 #define TCA_ACT_MAX_PRIO_MASK 0x1FF
617 int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
618 int nr_actions, struct tcf_result *res)
621 u32 jmp_ttl = TCA_ACT_MAX_PRIO; /*matches actions per filter */
625 if (skb_skip_tc_classify(skb))
629 for (i = 0; i < nr_actions; i++) {
630 const struct tc_action *a = actions[i];
632 if (jmp_prgcnt > 0) {
637 ret = a->ops->act(skb, a, res);
638 if (ret == TC_ACT_REPEAT)
639 goto repeat; /* we need a ttl - JHS */
641 if (TC_ACT_EXT_CMP(ret, TC_ACT_JUMP)) {
642 jmp_prgcnt = ret & TCA_ACT_MAX_PRIO_MASK;
643 if (!jmp_prgcnt || (jmp_prgcnt > nr_actions)) {
644 /* faulty opcode, stop pipeline */
649 goto restart_act_graph;
650 else /* faulty graph, stop pipeline */
653 } else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
654 tcf_action_goto_chain_exec(a, res);
657 if (ret != TC_ACT_PIPE)
663 EXPORT_SYMBOL(tcf_action_exec);
665 int tcf_action_destroy(struct tc_action *actions[], int bind)
667 const struct tc_action_ops *ops;
671 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
675 ret = __tcf_idr_release(a, bind, true);
676 if (ret == ACT_P_DELETED)
677 module_put(ops->owner);
684 static int tcf_action_put(struct tc_action *p)
686 return __tcf_action_put(p, false);
689 static void tcf_action_put_many(struct tc_action *actions[])
693 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
694 struct tc_action *a = actions[i];
695 const struct tc_action_ops *ops = a->ops;
697 if (tcf_action_put(a))
698 module_put(ops->owner);
703 tcf_action_dump_old(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
705 return a->ops->dump(skb, a, bind, ref);
709 tcf_action_dump_1(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
712 unsigned char *b = skb_tail_pointer(skb);
714 struct tc_cookie *cookie;
716 if (nla_put_string(skb, TCA_KIND, a->ops->kind))
717 goto nla_put_failure;
718 if (tcf_action_copy_stats(skb, a, 0))
719 goto nla_put_failure;
722 cookie = rcu_dereference(a->act_cookie);
724 if (nla_put(skb, TCA_ACT_COOKIE, cookie->len, cookie->data)) {
726 goto nla_put_failure;
731 nest = nla_nest_start(skb, TCA_OPTIONS);
733 goto nla_put_failure;
734 err = tcf_action_dump_old(skb, a, bind, ref);
736 nla_nest_end(skb, nest);
744 EXPORT_SYMBOL(tcf_action_dump_1);
746 int tcf_action_dump(struct sk_buff *skb, struct tc_action *actions[],
750 int err = -EINVAL, i;
753 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
755 nest = nla_nest_start(skb, a->order);
757 goto nla_put_failure;
758 err = tcf_action_dump_1(skb, a, bind, ref);
761 nla_nest_end(skb, nest);
769 nla_nest_cancel(skb, nest);
773 static struct tc_cookie *nla_memdup_cookie(struct nlattr **tb)
775 struct tc_cookie *c = kzalloc(sizeof(*c), GFP_KERNEL);
779 c->data = nla_memdup(tb[TCA_ACT_COOKIE], GFP_KERNEL);
784 c->len = nla_len(tb[TCA_ACT_COOKIE]);
789 static bool tcf_action_valid(int action)
791 int opcode = TC_ACT_EXT_OPCODE(action);
794 return action <= TC_ACT_VALUE_MAX;
795 return opcode <= TC_ACT_EXT_OPCODE_MAX || action == TC_ACT_UNSPEC;
798 struct tc_action *tcf_action_init_1(struct net *net, struct tcf_proto *tp,
799 struct nlattr *nla, struct nlattr *est,
800 char *name, int ovr, int bind,
802 struct netlink_ext_ack *extack)
805 struct tc_action_ops *a_o;
806 struct tc_cookie *cookie = NULL;
807 char act_name[IFNAMSIZ];
808 struct nlattr *tb[TCA_ACT_MAX + 1];
813 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
817 kind = tb[TCA_ACT_KIND];
819 NL_SET_ERR_MSG(extack, "TC action kind must be specified");
822 if (nla_strlcpy(act_name, kind, IFNAMSIZ) >= IFNAMSIZ) {
823 NL_SET_ERR_MSG(extack, "TC action name too long");
826 if (tb[TCA_ACT_COOKIE]) {
827 int cklen = nla_len(tb[TCA_ACT_COOKIE]);
829 if (cklen > TC_COOKIE_MAX_SIZE) {
830 NL_SET_ERR_MSG(extack, "TC cookie size above the maximum");
834 cookie = nla_memdup_cookie(tb);
836 NL_SET_ERR_MSG(extack, "No memory to generate TC cookie");
842 if (strlcpy(act_name, name, IFNAMSIZ) >= IFNAMSIZ) {
843 NL_SET_ERR_MSG(extack, "TC action name too long");
849 a_o = tc_lookup_action_n(act_name);
851 #ifdef CONFIG_MODULES
854 request_module("act_%s", act_name);
858 a_o = tc_lookup_action_n(act_name);
860 /* We dropped the RTNL semaphore in order to
861 * perform the module load. So, even if we
862 * succeeded in loading the module we have to
863 * tell the caller to replay the request. We
864 * indicate this using -EAGAIN.
871 NL_SET_ERR_MSG(extack, "Failed to load TC action module");
876 /* backward compatibility for policer */
878 err = a_o->init(net, tb[TCA_ACT_OPTIONS], est, &a, ovr, bind,
881 err = a_o->init(net, nla, est, &a, ovr, bind, rtnl_held,
886 if (!name && tb[TCA_ACT_COOKIE])
887 tcf_set_action_cookie(&a->act_cookie, cookie);
889 /* module count goes up only when brand new policy is created
890 * if it exists and is only bound to in a_o->init() then
891 * ACT_P_CREATED is not returned (a zero is).
893 if (err != ACT_P_CREATED)
894 module_put(a_o->owner);
896 if (TC_ACT_EXT_CMP(a->tcfa_action, TC_ACT_GOTO_CHAIN)) {
897 err = tcf_action_goto_chain_init(a, tp);
899 struct tc_action *actions[] = { a, NULL };
901 tcf_action_destroy(actions, bind);
902 NL_SET_ERR_MSG(extack, "Failed to init TC action chain");
907 if (!tcf_action_valid(a->tcfa_action)) {
908 NL_SET_ERR_MSG(extack, "invalid action value, using TC_ACT_UNSPEC instead");
909 a->tcfa_action = TC_ACT_UNSPEC;
915 module_put(a_o->owner);
924 /* Returns numbers of initialized actions or negative error. */
926 int tcf_action_init(struct net *net, struct tcf_proto *tp, struct nlattr *nla,
927 struct nlattr *est, char *name, int ovr, int bind,
928 struct tc_action *actions[], size_t *attr_size,
929 bool rtnl_held, struct netlink_ext_ack *extack)
931 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
932 struct tc_action *act;
937 err = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, extack);
941 for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
942 act = tcf_action_init_1(net, tp, tb[i], est, name, ovr, bind,
949 sz += tcf_action_fill_size(act);
950 /* Start from index 0 */
951 actions[i - 1] = act;
954 *attr_size = tcf_action_full_attrs_size(sz);
958 tcf_action_destroy(actions, bind);
962 int tcf_action_copy_stats(struct sk_buff *skb, struct tc_action *p,
971 /* compat_mode being true specifies a call that is supposed
972 * to add additional backward compatibility statistic TLVs.
975 if (p->type == TCA_OLD_COMPAT)
976 err = gnet_stats_start_copy_compat(skb, 0,
984 err = gnet_stats_start_copy(skb, TCA_ACT_STATS,
985 &p->tcfa_lock, &d, TCA_ACT_PAD);
990 if (gnet_stats_copy_basic(NULL, &d, p->cpu_bstats, &p->tcfa_bstats) < 0 ||
991 gnet_stats_copy_rate_est(&d, &p->tcfa_rate_est) < 0 ||
992 gnet_stats_copy_queue(&d, p->cpu_qstats,
994 p->tcfa_qstats.qlen) < 0)
997 if (gnet_stats_finish_copy(&d) < 0)
1006 static int tca_get_fill(struct sk_buff *skb, struct tc_action *actions[],
1007 u32 portid, u32 seq, u16 flags, int event, int bind,
1011 struct nlmsghdr *nlh;
1012 unsigned char *b = skb_tail_pointer(skb);
1013 struct nlattr *nest;
1015 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*t), flags);
1017 goto out_nlmsg_trim;
1018 t = nlmsg_data(nlh);
1019 t->tca_family = AF_UNSPEC;
1023 nest = nla_nest_start(skb, TCA_ACT_TAB);
1025 goto out_nlmsg_trim;
1027 if (tcf_action_dump(skb, actions, bind, ref) < 0)
1028 goto out_nlmsg_trim;
1030 nla_nest_end(skb, nest);
1032 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1041 tcf_get_notify(struct net *net, u32 portid, struct nlmsghdr *n,
1042 struct tc_action *actions[], int event,
1043 struct netlink_ext_ack *extack)
1045 struct sk_buff *skb;
1047 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1050 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, event,
1052 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1057 return rtnl_unicast(skb, net, portid);
1060 static struct tc_action *tcf_action_get_1(struct net *net, struct nlattr *nla,
1061 struct nlmsghdr *n, u32 portid,
1062 struct netlink_ext_ack *extack)
1064 struct nlattr *tb[TCA_ACT_MAX + 1];
1065 const struct tc_action_ops *ops;
1066 struct tc_action *a;
1070 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1075 if (tb[TCA_ACT_INDEX] == NULL ||
1076 nla_len(tb[TCA_ACT_INDEX]) < sizeof(index)) {
1077 NL_SET_ERR_MSG(extack, "Invalid TC action index value");
1080 index = nla_get_u32(tb[TCA_ACT_INDEX]);
1083 ops = tc_lookup_action(tb[TCA_ACT_KIND]);
1084 if (!ops) { /* could happen in batch of actions */
1085 NL_SET_ERR_MSG(extack, "Specified TC action not found");
1089 if (ops->lookup(net, &a, index, extack) == 0)
1092 module_put(ops->owner);
1096 module_put(ops->owner);
1098 return ERR_PTR(err);
1101 static int tca_action_flush(struct net *net, struct nlattr *nla,
1102 struct nlmsghdr *n, u32 portid,
1103 struct netlink_ext_ack *extack)
1105 struct sk_buff *skb;
1107 struct nlmsghdr *nlh;
1109 struct netlink_callback dcb;
1110 struct nlattr *nest;
1111 struct nlattr *tb[TCA_ACT_MAX + 1];
1112 const struct tc_action_ops *ops;
1113 struct nlattr *kind;
1116 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1120 b = skb_tail_pointer(skb);
1122 err = nla_parse_nested(tb, TCA_ACT_MAX, nla, NULL, extack);
1127 kind = tb[TCA_ACT_KIND];
1128 ops = tc_lookup_action(kind);
1129 if (!ops) { /*some idjot trying to flush unknown action */
1130 NL_SET_ERR_MSG(extack, "Cannot flush unknown TC action");
1134 nlh = nlmsg_put(skb, portid, n->nlmsg_seq, RTM_DELACTION,
1137 NL_SET_ERR_MSG(extack, "Failed to create TC action flush notification");
1138 goto out_module_put;
1140 t = nlmsg_data(nlh);
1141 t->tca_family = AF_UNSPEC;
1145 nest = nla_nest_start(skb, TCA_ACT_TAB);
1147 NL_SET_ERR_MSG(extack, "Failed to add new netlink message");
1148 goto out_module_put;
1151 err = ops->walk(net, skb, &dcb, RTM_DELACTION, ops, extack);
1153 nla_nest_cancel(skb, nest);
1154 goto out_module_put;
1157 nla_nest_end(skb, nest);
1159 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1160 nlh->nlmsg_flags |= NLM_F_ROOT;
1161 module_put(ops->owner);
1162 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1163 n->nlmsg_flags & NLM_F_ECHO);
1167 NL_SET_ERR_MSG(extack, "Failed to send TC action flush notification");
1172 module_put(ops->owner);
1178 static int tcf_action_delete(struct net *net, struct tc_action *actions[],
1179 int *acts_deleted, struct netlink_ext_ack *extack)
1184 for (i = 0; i < TCA_ACT_MAX_PRIO && actions[i]; i++) {
1185 struct tc_action *a = actions[i];
1186 const struct tc_action_ops *ops = a->ops;
1188 /* Actions can be deleted concurrently so we must save their
1189 * type and id to search again after reference is released.
1191 act_index = a->tcfa_index;
1193 if (tcf_action_put(a)) {
1194 /* last reference, action was deleted concurrently */
1195 module_put(ops->owner);
1197 /* now do the delete */
1198 ret = ops->delete(net, act_index);
1200 *acts_deleted = i + 1;
1210 tcf_del_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1211 int *acts_deleted, u32 portid, size_t attr_size,
1212 struct netlink_ext_ack *extack)
1215 struct sk_buff *skb;
1217 skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1222 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, 0, RTM_DELACTION,
1224 NL_SET_ERR_MSG(extack, "Failed to fill netlink TC action attributes");
1229 /* now do the delete */
1230 ret = tcf_action_delete(net, actions, acts_deleted, extack);
1232 NL_SET_ERR_MSG(extack, "Failed to delete TC action");
1237 ret = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1238 n->nlmsg_flags & NLM_F_ECHO);
1245 tca_action_gd(struct net *net, struct nlattr *nla, struct nlmsghdr *n,
1246 u32 portid, int event, struct netlink_ext_ack *extack)
1249 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1250 struct tc_action *act;
1251 size_t attr_size = 0;
1252 struct tc_action *actions[TCA_ACT_MAX_PRIO + 1] = {};
1253 int acts_deleted = 0;
1255 ret = nla_parse_nested(tb, TCA_ACT_MAX_PRIO, nla, NULL, extack);
1259 if (event == RTM_DELACTION && n->nlmsg_flags & NLM_F_ROOT) {
1261 return tca_action_flush(net, tb[1], n, portid, extack);
1263 NL_SET_ERR_MSG(extack, "Invalid netlink attributes while flushing TC action");
1267 for (i = 1; i <= TCA_ACT_MAX_PRIO && tb[i]; i++) {
1268 act = tcf_action_get_1(net, tb[i], n, portid, extack);
1274 attr_size += tcf_action_fill_size(act);
1275 actions[i - 1] = act;
1278 attr_size = tcf_action_full_attrs_size(attr_size);
1280 if (event == RTM_GETACTION)
1281 ret = tcf_get_notify(net, portid, n, actions, event, extack);
1283 ret = tcf_del_notify(net, n, actions, &acts_deleted, portid,
1290 tcf_action_put_many(&actions[acts_deleted]);
1295 tcf_add_notify(struct net *net, struct nlmsghdr *n, struct tc_action *actions[],
1296 u32 portid, size_t attr_size, struct netlink_ext_ack *extack)
1298 struct sk_buff *skb;
1301 skb = alloc_skb(attr_size <= NLMSG_GOODSIZE ? NLMSG_GOODSIZE : attr_size,
1306 if (tca_get_fill(skb, actions, portid, n->nlmsg_seq, n->nlmsg_flags,
1307 RTM_NEWACTION, 0, 0) <= 0) {
1308 NL_SET_ERR_MSG(extack, "Failed to fill netlink attributes while adding TC action");
1313 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1314 n->nlmsg_flags & NLM_F_ECHO);
1320 static int tcf_action_add(struct net *net, struct nlattr *nla,
1321 struct nlmsghdr *n, u32 portid, int ovr,
1322 struct netlink_ext_ack *extack)
1324 size_t attr_size = 0;
1326 struct tc_action *actions[TCA_ACT_MAX_PRIO] = {};
1328 ret = tcf_action_init(net, NULL, nla, NULL, NULL, ovr, 0, actions,
1329 &attr_size, true, extack);
1332 ret = tcf_add_notify(net, n, actions, portid, attr_size, extack);
1334 tcf_action_put_many(actions);
1339 static u32 tcaa_root_flags_allowed = TCA_FLAG_LARGE_DUMP_ON;
1340 static const struct nla_policy tcaa_policy[TCA_ROOT_MAX + 1] = {
1341 [TCA_ROOT_FLAGS] = { .type = NLA_BITFIELD32,
1342 .validation_data = &tcaa_root_flags_allowed },
1343 [TCA_ROOT_TIME_DELTA] = { .type = NLA_U32 },
1346 static int tc_ctl_action(struct sk_buff *skb, struct nlmsghdr *n,
1347 struct netlink_ext_ack *extack)
1349 struct net *net = sock_net(skb->sk);
1350 struct nlattr *tca[TCA_ROOT_MAX + 1];
1351 u32 portid = skb ? NETLINK_CB(skb).portid : 0;
1352 int ret = 0, ovr = 0;
1354 if ((n->nlmsg_type != RTM_GETACTION) &&
1355 !netlink_capable(skb, CAP_NET_ADMIN))
1358 ret = nlmsg_parse(n, sizeof(struct tcamsg), tca, TCA_ROOT_MAX, NULL,
1363 if (tca[TCA_ACT_TAB] == NULL) {
1364 NL_SET_ERR_MSG(extack, "Netlink action attributes missing");
1368 /* n->nlmsg_flags & NLM_F_CREATE */
1369 switch (n->nlmsg_type) {
1371 /* we are going to assume all other flags
1372 * imply create only if it doesn't exist
1373 * Note that CREATE | EXCL implies that
1374 * but since we want avoid ambiguity (eg when flags
1375 * is zero) then just set this
1377 if (n->nlmsg_flags & NLM_F_REPLACE)
1380 ret = tcf_action_add(net, tca[TCA_ACT_TAB], n, portid, ovr,
1386 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1387 portid, RTM_DELACTION, extack);
1390 ret = tca_action_gd(net, tca[TCA_ACT_TAB], n,
1391 portid, RTM_GETACTION, extack);
1400 static struct nlattr *find_dump_kind(struct nlattr **nla)
1402 struct nlattr *tb1, *tb2[TCA_ACT_MAX + 1];
1403 struct nlattr *tb[TCA_ACT_MAX_PRIO + 1];
1404 struct nlattr *kind;
1406 tb1 = nla[TCA_ACT_TAB];
1410 if (nla_parse(tb, TCA_ACT_MAX_PRIO, nla_data(tb1),
1411 NLMSG_ALIGN(nla_len(tb1)), NULL, NULL) < 0)
1416 if (nla_parse_nested(tb2, TCA_ACT_MAX, tb[1], NULL, NULL) < 0)
1418 kind = tb2[TCA_ACT_KIND];
1423 static int tc_dump_action(struct sk_buff *skb, struct netlink_callback *cb)
1425 struct net *net = sock_net(skb->sk);
1426 struct nlmsghdr *nlh;
1427 unsigned char *b = skb_tail_pointer(skb);
1428 struct nlattr *nest;
1429 struct tc_action_ops *a_o;
1431 struct tcamsg *t = (struct tcamsg *) nlmsg_data(cb->nlh);
1432 struct nlattr *tb[TCA_ROOT_MAX + 1];
1433 struct nlattr *count_attr = NULL;
1434 unsigned long jiffy_since = 0;
1435 struct nlattr *kind = NULL;
1436 struct nla_bitfield32 bf;
1437 u32 msecs_since = 0;
1440 ret = nlmsg_parse(cb->nlh, sizeof(struct tcamsg), tb, TCA_ROOT_MAX,
1445 kind = find_dump_kind(tb);
1447 pr_info("tc_dump_action: action bad kind\n");
1451 a_o = tc_lookup_action(kind);
1456 if (tb[TCA_ROOT_FLAGS]) {
1457 bf = nla_get_bitfield32(tb[TCA_ROOT_FLAGS]);
1458 cb->args[2] = bf.value;
1461 if (tb[TCA_ROOT_TIME_DELTA]) {
1462 msecs_since = nla_get_u32(tb[TCA_ROOT_TIME_DELTA]);
1465 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1466 cb->nlh->nlmsg_type, sizeof(*t), 0);
1468 goto out_module_put;
1471 jiffy_since = jiffies - msecs_to_jiffies(msecs_since);
1473 t = nlmsg_data(nlh);
1474 t->tca_family = AF_UNSPEC;
1477 cb->args[3] = jiffy_since;
1478 count_attr = nla_reserve(skb, TCA_ROOT_COUNT, sizeof(u32));
1480 goto out_module_put;
1482 nest = nla_nest_start(skb, TCA_ACT_TAB);
1484 goto out_module_put;
1486 ret = a_o->walk(net, skb, cb, RTM_GETACTION, a_o, NULL);
1488 goto out_module_put;
1491 nla_nest_end(skb, nest);
1493 act_count = cb->args[1];
1494 memcpy(nla_data(count_attr), &act_count, sizeof(u32));
1499 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1500 if (NETLINK_CB(cb->skb).portid && ret)
1501 nlh->nlmsg_flags |= NLM_F_MULTI;
1502 module_put(a_o->owner);
1506 module_put(a_o->owner);
1511 struct tcf_action_net {
1512 struct rhashtable egdev_ht;
1515 static unsigned int tcf_action_net_id;
1517 struct tcf_action_egdev_cb {
1518 struct list_head list;
1523 struct tcf_action_egdev {
1524 struct rhash_head ht_node;
1525 const struct net_device *dev;
1526 unsigned int refcnt;
1527 struct list_head cb_list;
1530 static const struct rhashtable_params tcf_action_egdev_ht_params = {
1531 .key_offset = offsetof(struct tcf_action_egdev, dev),
1532 .head_offset = offsetof(struct tcf_action_egdev, ht_node),
1533 .key_len = sizeof(const struct net_device *),
1536 static struct tcf_action_egdev *
1537 tcf_action_egdev_lookup(const struct net_device *dev)
1539 struct net *net = dev_net(dev);
1540 struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1542 return rhashtable_lookup_fast(&tan->egdev_ht, &dev,
1543 tcf_action_egdev_ht_params);
1546 static struct tcf_action_egdev *
1547 tcf_action_egdev_get(const struct net_device *dev)
1549 struct tcf_action_egdev *egdev;
1550 struct tcf_action_net *tan;
1552 egdev = tcf_action_egdev_lookup(dev);
1556 egdev = kzalloc(sizeof(*egdev), GFP_KERNEL);
1559 INIT_LIST_HEAD(&egdev->cb_list);
1561 tan = net_generic(dev_net(dev), tcf_action_net_id);
1562 rhashtable_insert_fast(&tan->egdev_ht, &egdev->ht_node,
1563 tcf_action_egdev_ht_params);
1570 static void tcf_action_egdev_put(struct tcf_action_egdev *egdev)
1572 struct tcf_action_net *tan;
1574 if (--egdev->refcnt)
1576 tan = net_generic(dev_net(egdev->dev), tcf_action_net_id);
1577 rhashtable_remove_fast(&tan->egdev_ht, &egdev->ht_node,
1578 tcf_action_egdev_ht_params);
1582 static struct tcf_action_egdev_cb *
1583 tcf_action_egdev_cb_lookup(struct tcf_action_egdev *egdev,
1584 tc_setup_cb_t *cb, void *cb_priv)
1586 struct tcf_action_egdev_cb *egdev_cb;
1588 list_for_each_entry(egdev_cb, &egdev->cb_list, list)
1589 if (egdev_cb->cb == cb && egdev_cb->cb_priv == cb_priv)
1594 static int tcf_action_egdev_cb_call(struct tcf_action_egdev *egdev,
1595 enum tc_setup_type type,
1596 void *type_data, bool err_stop)
1598 struct tcf_action_egdev_cb *egdev_cb;
1602 list_for_each_entry(egdev_cb, &egdev->cb_list, list) {
1603 err = egdev_cb->cb(type, type_data, egdev_cb->cb_priv);
1614 static int tcf_action_egdev_cb_add(struct tcf_action_egdev *egdev,
1615 tc_setup_cb_t *cb, void *cb_priv)
1617 struct tcf_action_egdev_cb *egdev_cb;
1619 egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1620 if (WARN_ON(egdev_cb))
1622 egdev_cb = kzalloc(sizeof(*egdev_cb), GFP_KERNEL);
1626 egdev_cb->cb_priv = cb_priv;
1627 list_add(&egdev_cb->list, &egdev->cb_list);
1631 static void tcf_action_egdev_cb_del(struct tcf_action_egdev *egdev,
1632 tc_setup_cb_t *cb, void *cb_priv)
1634 struct tcf_action_egdev_cb *egdev_cb;
1636 egdev_cb = tcf_action_egdev_cb_lookup(egdev, cb, cb_priv);
1637 if (WARN_ON(!egdev_cb))
1639 list_del(&egdev_cb->list);
1643 static int __tc_setup_cb_egdev_register(const struct net_device *dev,
1644 tc_setup_cb_t *cb, void *cb_priv)
1646 struct tcf_action_egdev *egdev = tcf_action_egdev_get(dev);
1651 err = tcf_action_egdev_cb_add(egdev, cb, cb_priv);
1657 tcf_action_egdev_put(egdev);
1660 int tc_setup_cb_egdev_register(const struct net_device *dev,
1661 tc_setup_cb_t *cb, void *cb_priv)
1666 err = __tc_setup_cb_egdev_register(dev, cb, cb_priv);
1670 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_register);
1672 static void __tc_setup_cb_egdev_unregister(const struct net_device *dev,
1673 tc_setup_cb_t *cb, void *cb_priv)
1675 struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1677 if (WARN_ON(!egdev))
1679 tcf_action_egdev_cb_del(egdev, cb, cb_priv);
1680 tcf_action_egdev_put(egdev);
1682 void tc_setup_cb_egdev_unregister(const struct net_device *dev,
1683 tc_setup_cb_t *cb, void *cb_priv)
1686 __tc_setup_cb_egdev_unregister(dev, cb, cb_priv);
1689 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_unregister);
1691 int tc_setup_cb_egdev_call(const struct net_device *dev,
1692 enum tc_setup_type type, void *type_data,
1695 struct tcf_action_egdev *egdev = tcf_action_egdev_lookup(dev);
1699 return tcf_action_egdev_cb_call(egdev, type, type_data, err_stop);
1701 EXPORT_SYMBOL_GPL(tc_setup_cb_egdev_call);
1703 static __net_init int tcf_action_net_init(struct net *net)
1705 struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1707 return rhashtable_init(&tan->egdev_ht, &tcf_action_egdev_ht_params);
1710 static void __net_exit tcf_action_net_exit(struct net *net)
1712 struct tcf_action_net *tan = net_generic(net, tcf_action_net_id);
1714 rhashtable_destroy(&tan->egdev_ht);
1717 static struct pernet_operations tcf_action_net_ops = {
1718 .init = tcf_action_net_init,
1719 .exit = tcf_action_net_exit,
1720 .id = &tcf_action_net_id,
1721 .size = sizeof(struct tcf_action_net),
1724 static int __init tc_action_init(void)
1728 err = register_pernet_subsys(&tcf_action_net_ops);
1732 rtnl_register(PF_UNSPEC, RTM_NEWACTION, tc_ctl_action, NULL, 0);
1733 rtnl_register(PF_UNSPEC, RTM_DELACTION, tc_ctl_action, NULL, 0);
1734 rtnl_register(PF_UNSPEC, RTM_GETACTION, tc_ctl_action, tc_dump_action,
1740 subsys_initcall(tc_action_init);